Physical Sciences and Mathematics Commons^™

Image Restoration Using Automatic Damaged Regions Detection And Machine Learning-Based Inpainting Technique, Chloe Martin-King

Computational and Data Sciences (PhD) Dissertations

In this dissertation we propose two novel image restoration schemes. The first pertains to automatic detection of damaged regions in old photographs and digital images of cracked paintings. In cases when inpainting mask generation cannot be completely automatic, our detection algorithm facilitates precise mask creation, particularly useful for images containing damage that is tedious to annotate or difficult to geometrically define. The main contribution of this dissertation is the development and utilization of a new inpainting technique, region hiding, to repair a single image by training a convolutional neural network on various transformations of that image. Region hiding is also …

Detecting Digitally Forged Faces In Online Videos, Neilesh Sambhu

USF Tampa Graduate Theses and Dissertations

We use Rossler’s FaceForensics dataset of 1004 online videos and their corresponding forged counterparts [1] to investigate the ability to distinguish digitally forged facial images from original images automatically with deep learning. The proposed convolutional neural network is much smaller than the current state-of-the-art solutions. Nevertheless, the network maintains a high level of accuracy (99.6%), all while using the entire FaceForensics dataset and not including any temporal information. We implement majority voting and show the impact on accuracy (99.67%), where only 1 video of 300 is misclassified. We examine why the model misclassified this one video. In terms of tuning …

Incorporating Figure Captions And Descriptive Text Into Mesh Term Indexing: A Deep Learning Approach, Xindi Wang

Electronic Thesis and Dissertation Repository

The exponential increase of available documents online makes document classification an important application in natural language processing. The goal of text classification is to automatically assign categories to documents. Traditional text classifiers depend on features, such as, vocabulary and user-specified information which mainly relies on prior knowledge. In contrast, deep learning automatically learns effective features from data instead of adopting human-designed features. In this thesis, we specifically focus on biomedical document classification. Beyond text information from abstract and title, we also consider image and table captions, as well as paragraphs associated with images and tables, which we demonstrate to be …

Fault Diagnosis Of High Speed Train Bogie Based On Multi-Domain Fusion Cnn, Yunpu Wu, Weidong Jin, Yingkun Huang

Journal of System Simulation

Abstract: The performance degradation and failures of high-speed train bogie components directly threaten the operation security of train. A fault detection method based on multi-domain fusion convolutional neural network is proposed to address the high complexity, high coupling and strong nonlinearity of vibration signals. Noise injection for time domain signal is used to enhance noise robustness and generalization of the model. Signal time-frequency representation information is obtained through embedded time-frequency transformation layer. Adaptive weight-based fusion is implemented through intrinsic characteristics of the convolutional neural network to handle the multi-domain multi-channel information. The experimental results show that the proposed method improves …

Cloud Fraction Of Satellite Imagery Based On Convolutional Neural Networks, Xia Min, Maoyang Shen, Jianfeng Wang, Yangguang Wang

Journal of System Simulation

Abstract: Cloud fraction is the basis for the application of meteorological satellite. Existing methods cannot use all the characteristics and optical parameters of the satellite cloud, which results in the inaccuracy of cloud detection and cloud fraction. In order to solve this problem, convolutional neural network is used for cloud detection. Based on the improved convolutional neural network, the satellite cloud image is divided into thin cloud, thick cloud and clear sky. Based on the cloud detection, an improved spatial correlation method is used for cloud fraction. The results for Chinese HJ-1A/B satellite imagery show that convolutional neural network can …

End-To-End Learning Via A Convolutional Neural Network For Cancer Cell Line Classification, Darlington A. Akogo, Xavier-Lewis Palmer

Electrical & Computer Engineering Faculty Publications

Purpose: Computer vision for automated analysis of cells and tissues usually include extracting features from images before analyzing such features via various machine learning and machine vision algorithms. The purpose of this work is to explore and demonstrate the ability of a Convolutional Neural Network (CNN) to classify cells pictured via brightfield microscopy without the need of any feature extraction, using a minimum of images, improving work-flows that involve cancer cell identification.

Design/methodology/approach: The methodology involved a quantitative measure of the performance of a Convolutional Neural Network in distinguishing between two cancer lines. In their approach, they trained, validated and …